Multidimensional scaling of quality judgments of speech signals processed by hearing aids

Perceptual dimensions underlying similarity ratings and preference judgments of the quality of hearing-aid-processed speech were derived via multidimensional-scaling procedures, and were correlated with 15 indices of electroacustic performance to determine those electroacoustic characteristics contributing maximally to quality judgements. Connected discourse was tape recorded through 20 conventional hearing aids and presented in a paired-comparison paradigm to ten normal listeners for similarity ratings and quality preference judgments. Analysis of the group similarity matrices revealed only one definitive perceptual dimension, low-cutoff frequency, that was common to the listeners. In a three-dimensional INDSCAL solution, this electroacoustic factor correlated--0.87 with dimension 1. Listeners strongly preferred hearing aids with relatively low low-cutoff frequencies; dimension 1 correlated--0.76 with their preferences. The quality judgments of some listeners were found via KYST analysis to exhibit a multidimensional structure, with low-cutoff frequency remaining the most salient dimension. Secondary dimensions varied across these several listeners, and included response irregularity, equivalent input noise level, high-cutoff frequency, and intermodulation distortion. Thus, the data provided evidence that, while low-cutoff frequency dominates listener judgments a multidimensional model may underlie the perceptual strategies used by some listeners to form quality judgments of hearing-aid-processed speech.